Quantifying memory CD8 T cell-mediated immune pressure on influenza A virus infection in vivo

The conservation of T cell epitopes in human influenza A virus has prompted the development of T cell-inducing influenza vaccines. However, the selection pressure mediated by memory CD8 T cells upon influenza virus has not been directly measured. Using a droplet digital PCR technique to distinguish wild-type and an epitope-mutant PR8 influenza viruses in vivo, this study quantifies the viral replicative fitness of a CD8 T cell-escaping mutation in the immunodominant influenza NP366-374 epitope in C57BL/6 (B6) mice under different settings of cellular immunity. Although this mutation does not result in a viral fitness defect in vitro or during the early stages of in vivo infection in naïve B6 mice, it does confer a moderate but consistent advantage to the mutant virus following heterosubtypic challenge of HKx31-immunized mice. In addition, this advantage was maintained under increased MHC diversity but became more substantial when the breadth of epitope recognition is limited. Finally, we showed that lung-resident, but not circulating, memory CD8 T cells are the primary source of cellular immune pressure early during infection, prior to the induction of a secondary effector T cell response. Integrating the data with an established modeling framework, we show that the relatively modest immune pressure mediated by memory CD8 T cells is one of the important factors responsible for the conservation of CD8 T cell epitopes in influenza A viruses that circulate among humans. Thus, a T cell-inducing vaccine that generates lung-resident memory CD8 T cells covering a sufficient breadth of epitopes may transiently protect against severe pathology without driving the virus to rapidly evolve and escape. Author Summary Since the historic Spanish flu in 1918, influenza has caused several pandemics and become an important public health concern. The inactivated vaccines routinely used attempt to boost antibodies, which may not be as effective when antigenic mismatch happens and could drive the virus to evolve and escape due to their high immune pressure. In contrast, the ability of influenza-specific T cells to reduce pathology and the conservation of T-cell epitopes across subtypes have shed light on the development of universal vaccines. In this study, we assessed the CD8 T cell-mediated selection pressure on influenza virus in mouse using a digital PCR technique. Within mice that have influenza-specific systemic and lung-resident memory CD8 T cells established, we found the advantage conferred by an escaping mutation in one of the immunodominant epitopes is around 25%. This advantage becomes much greater when the cellular immunity focuses on the focal epitope, while it is delayed when only systemic cellular immunity is established. Combining the data with our previous modeling work, we conclude that the small selection pressure imposed by CD8 T cells can explain the overall conservation of CD8 T cell epitopes of influenza A virus in addition to functional constraint.


Introduction
Influenza is an important public health issue, resulting in approximately 410,000 deaths worldwide annually [1]. The current inactivated influenza vaccines aim to induce antibodies that prevent viral infection; however, this strategy has two major stipulations: (i) an effective antibody response requires the match of antigenicity between circulating and vaccine strains [2,3], and (ii) the selection pressure mediated by neutralizing antibodies may drive the virus to evolve and escape [4,5].
Memory CD8 T cells are categorized into central (TCM), effector (TEM), and tissue-resident (TRM), according to their ability to circulate between secondary lymphoid organs, blood, and peripheral tissues [11][12][13]. Although influenza-specific memory T cells are not likely to provide sterilizing immunity, CD8 T cell responses, lung TRM in particular, have been shown to prevent severe pathology in mice [14][15][16][17][18] and reduce symptoms in humans [19] following heterosubtypic influenza infection. Although it hasn't been measured directly, the ability of antiviral memory CD8 T cells to limit influenza virus replication may also limit viral transmission [19,20]. The ability of CD8 T cell to recognize and respond to diverse influenza strains and subtypes is mainly due to the conservation of CD8 T cell epitopes. Out of 64 known CD8 T cell NP epitopes across all human leukocyte antigen (HLA) alleles, only 6 epitopes have been experimentally verified to have escaping mutations [21], underpinning the potential of developing a T cell-inducing vaccine to combat influenza.
Several studies have attempted to elucidate the reason why influenza virus epitopes recognized by CD8 T cells are conserved despite cellular immune pressure. One of the mainstream hypotheses is that detrimental fitness effects would result from mutations to viral proteins that must interact with a range of host proteins to ensure efficient replication [22][23][24]. Although epistatic interactions with compensatory mutations may recover the viral fitness [25,26], this process may slow the rate of generating competitive mutants. In our previous modeling work, we hypothesized that small selection pressure imposed by CD8 T cells, combined with human MHC polymorphisms, may further limit the rate of invasion of an escaping mutant [21]. Viral evolution due to CD8 T cell immunity has been observed in HIV [27,28], HCV [29], and chronic influenza infection of immunocompromised patients [30]. However, quantitative measures of the selection pressure from memory CD8 T cells on influenza viral dynamics in vivo are lacking in the context of influenza infections in immunocompetent hosts.
In this study, we aim to quantify the selective advantage of a CD8 T cell-escaping influenza mutant under different contexts related to (i) distinct memory CD8 T cell subsets and (ii) the breadth of memory CD8 T cell responses. The mouse-adapted influenza strain, A/Puerto Rico/8/1934 (H1N1) (PR8), harbors three H-2 b -restricted immunodominant CD8 T cell epitopes [31,32]. We used the wild-type PR8 and PR8 with a point mutation on one of the immunodominant epitopes, NP366-374, such that the peptide-MHCI binding is disrupted and the mutant epitope is not presented [33]. To minimize the between-subject variation in measuring the viral kinetics, we used a droplet digital PCR system to simultaneously measure the viral loads of both wild-type and epitope-mutant viruses in a single co-infected animal at selected time points. Then, the selection coefficient of the escaping mutant was estimated by comparing the area under growth curves of the two viruses. Integrating these data with our modeling framework allowed us to quantify the factors that determine the rate of invasion of epitope-variants that escape CD8 T cell responses.
Our results are consistent with the argument that broad memory CD8 T cell immunity imposes modest selection pressure on the virus, which helps explain the overall conservation of CD8 T cell epitopes of human IAV.

In vitro viral growth detects no significant fitness defect of NP-N370Q mutation
Prior studies have identified a mutation in the immunodominant PR8 influenza nucleoprotein (NP)366-374 CD8 T cell epitope presented on H-2 b that prevents loading of the peptide onto MHCI [33]. We constructed wild-type (WT) and NP-N370Q mutant (MT) PR8 influenza viruses by reverse genetics and first compared viral growth in MDCK cell culture to determine if this mutation impacted viral fitness. The viral titers at five time points after inoculation are shown in Fig 1, where no significant difference was detected between virus strains (three-way ANOVA, p-value for virus strains = 0.797). A further logistic growth model fitting found no significant difference between the estimated model parameters associated with WT and MT PR8 viruses (Table 1).
Therefore, we conclude that the NP-N370Q mutation does not result in a fitness defect in vitro.

Droplet digital PCR robustly differentiates between the wild type and mutant
In order to infer the impact of CD8 T cell-escaping mutations from in vivo experiments, we first had to design and validate a method for simultaneous measurement of WT and MT PR8 viruses within the same host. Thus, we employed a droplet digital PCR (ddPCR) system with probes specific for the WT or MT variants of the NP366-374 epitope (See Supplemental Information). The assay is highly specific for the individual strains (Fig 2A), as the WT probe detected positive signals only from the WT viral RNA samples, and the MT probe was similarly specific for the MT viral RNA samples. This assay also robustly reflects the change in RNA concentration, as a 10-fold dilution of input viruses resulted in a 10-fold decrease in the final readouts indicating unbiased detection of WT and MT viruses. (Fig 2B). Furthermore, the viral loads measured by ddPCR (in copies/mL) were consistent with viral titers measured by plaque assay (in pfu/mL), and there was no preferential detection of WT or MT viral RNA ( Table 2). Thus, the ddPCR system allows us to accurately differentiate between WT and MT PR8 viruses in a mixed sample.

Table 2
Comparison of viral load measured by droplet digital PCR (copy number/mL) and viral titer measured by plaque assay (pfu/mL).

Wild type and mutant viruses have similar fitness during the early stage of primary infection
To determine if the MT PR8 virus has any fitness defect in vivo in the absence of influenza-specific memory CD8 T cell immunity, we infected naïve C57BL/6 (B6) mice with an equal mixture of WT and MT PR8 viruses (Fig 3A). In naïve B6 mice, the viral loads of WT and MT viruses in the lungs recapitulated the kinetics reported previously [34], where they (i) increased exponentially and peaked around 3 days post-infection (dpi), (ii) slowly decayed from 3 to 5 dpi, and (iii) quickly decayed and were cleared after 5 dpi (Fig 3B). Across the 5 time points being sampled, only on day 5 was the average log-transformed viral load of MT significantly higher than the WT (p = 0.0007).
We then defined the selection coefficient of MT as The selection coefficient of MT gradually increased from 0.001 on day 1 to 0.1 on day 9, while the 99% empirical bootstrap confidence interval (99% CI hereafter) did not contain zero only on day 9 (Fig 3C). These findings showed that the MT virus does not have a detectable fitness defect during the early stages of infection in vivo, but gains an advantage as the infection progresses, likely due to its ability to escape from the immunodominant NP366-374-specific effector CD8 T cells, similar to previous reports [34].

The mutant virus acquires selective advantage early during secondary heterosubtypic infection in the presence of influenza-specific memory CD8 T cell immunity
To estimate the selective advantage of the MT virus under different settings of cellular immunity, we intranasally (i.n.) primed B6 mice with HKx31 (H3N2), challenged the mice 30 days later with an equal mixture of WT and MT PR8 (H1N1) viruses, and measured the viral kinetics (Fig 4A).
Viral loads peaked on day 2, slowly decayed from days 2 to 4, and rapidly decayed and cleared from days 4 to 8. Despite a similar overall pattern to naïve mice, the peak viral loads were around 10-fold lower, and clearance was faster. The means of log-transformed MT viral loads were significantly higher than those of WT on days 1 and 4 (p = 0.034 and 0.0018, respectively) (Fig 4B).
Likewise, the selection coefficient of MT continuously increased from 0.15 on day 1 to 0.27 on day 9, while the 99% CIs contained zero only on day 1 (Fig 4C). Thus, the MT acquired an advantage in the early stage of infection, and the advantage persisted and increased through the infection course.
To investigate whether an increased MHC diversity impacts the selective advantage of the MT virus, we applied the same prime-challenge procedure to CB6F1 (F1) mice, which are the offspring of C57BL/6 and BALB/c mouse strains, and harbor both H-2 d and H-2 b haplotypes ( Fig   4D). Thus, F1 mice develop a broader influenza-specific CD8 T cell response that encompasses epitopes presented by both MHC alleles. We observed similar viral kinetics in F1 mice compared to B6 mice; however, the viruses were cleared even faster than B6 mice (no virus was detected on days 6 and 8), and the MT significantly outgrew WT only on day 4 (p = 0.0054) (Fig 4E).
When looking at the selection coefficient of the MT virus, we noticed two interesting differences between i.n.-primed B6 and F1 mice (Fig 4F) instead, it depends on the time of observation.

CD8 T cell-escaping mutations confer a much greater advantage to the MT virus when the breadth of CD8 T cell response is limited
Intranasal priming of B6 mice with x31 generates memory CD8 T cells specific for multiple epitopes in addition to the immunodominant NP366-374 epitope, including two additional immunodominant and at least eight subdominant epitopes [31,32]. Therefore, even if the MT PR8 virus escapes detection from NP366-374-specific CD8 T cells, it remains subject to recognition by CD8 T cells targeting other epitopes. However, some vaccine approaches are designed to focus the CD8 T cell response to a few immunodominant epitopes, raising the question of the selective advantage that can be gained by a mutant influenza virus that can escape a focused influenza-specific memory CD8 T cell repertoire. To address this question, we immunized B6 mice with a recombinant, replication-deficient adenovirus 5 expressing the PR8 influenza nucleoprotein (AdNP), which generates memory CD8 T cells specific for only NP-derived epitopes, mainly the immunodominant NP366-374 epitope [35], and challenged them 30 days later with a mixture of WT and MT PR8 viruses (Fig 5A). Vastly different growth kinetics of the MT virus were observed compared to the WT virus (Fig 5B). The MT grew continuously and peaked around day 4, at a viral load 35-fold higher than WT. After day 4, both viruses decayed at the same rate exponentially, but neither were completely cleared on day 8. At each time point selected, the means of log-transformed MT viral loads were significantly higher than those of WT (p = 0.031, 0.0004, 0.0002, 0.0016, and 0.023 for days 1, 2, 4, 6, and 8, respectively). Consistent with this, the selection coefficient of MT continuously climbed throughout the infectious course from 0.24 on day 1 to 0.8 on day 8, and all the 99% CIs did not contain zero (Fig 5C). These data show that when the antiviral memory CD8 T cell repertoire is comprised of only a single epitope, the selective advantage conferred by the corresponding escaping mutation is substantially high, much greater than in the context where memory CD8 T cells recognize multiple epitopes.

Lung-resident memory CD8 T cells are the primary source of selection pressure during the early stage of secondary infection
Lung-resident memory CD8 T cells (lung TRM) are important for mediating heterosubtypic immunity and controlling influenza virus replication due to their localization in the respiratory tract enabling rapid detection of infected cells. Thus, we hypothesize lung TRM impose much of the early cellular immune pressure on the virus. We tested this hypothesis by comparing the viral kinetics in 30-day i.n. x31-primed B6 mice with 30-day intramuscularly (i.m.) x31-primed B6 mice (Fig 6A), which we previously showed to have approximately the same number of influenzaspecific systemic memory CD8 T cells as i.n.-primed mice but lack lung TRM [17]. We observed that, in i.m. x31-primed mice, (i) the viral loads peaked at a later time point (day 4) and were about 6-and 12-fold higher than the peaks seen in i.n. x31-primed mice, (ii) the MT and WT viruses grew at the same rate between days 1-4 and peaked at similar levels, and (iii) after day 4, the WT virus was cleared faster than the MT virus (p = 0.021 on day 6, p = 0.0095 on day 8) (Fig 6B).
Interestingly, the selection coefficient of MT showed a U-shape trend; it decreased from 0.12 on day 1 to 0.057 on day 4, and then increased to 0.22 on day 8 (Fig 6C). Nevertheless, these data showed the MT virus does not acquire the same increased advantage during the early stage of infection when the lung TRM is absent. Thus, the selection pressure mediated by lung TRM is the main driver of the outgrowth of MT virus observed in i.n. x31-primed B6 mice.

Comparison of the selection coefficients of MT virus revealed how cellular immunity impacts the selective advantage acquired by escaping mutations
We summarized the selection coefficients of MT virus estimated from log-transformed viral kinetics in Table 3. Overall, the MT had higher fitness than the WT through the infection course, regardless of the context of pre-existing cellular immunity. The fitness gain ranged from 10% to 74%, depending on the immune settings. However, a stratifying analysis revealed that the fitness gain does not evenly distribute across the infection course. During the first 4 or 5 days, the MT had little advantage in naïve and i.m. x31-primed mice. This implied most of the advantage was acquired later during infection under these scenarios. In contrast, the MT acquired around 15% increase in fitness in both i.n. x31-primed B6 and F1 mice through days 0-4, but over the whole infection course this advantage increased to 24% in B6 mice while it was maintained in F1, corresponding to faster viral clearance in F1 mice. Lastly, the MT acquired a large and stable fitness gain in AdNP-primed B6 mice, at an average rate 8% per day. In summary, the NP-N370Q mutation confers no more than 25% increase in fitness when CD8 T cell immunity targets multiple epitopes, while it can confer up to 74% increase in fitness when only the NP366-374 epitope is targeted.
We explored the reason why escape from NP366-374-specific CD8 T cells only confers modest selective advantage, focusing on CD8 T cell kinetics during secondary infection. We challenged i.n. x31-primed B6 mice with either WT or MT viruses alone and tracked the numbers and percentages of CD44 + Tetramer + CD8 T cells in the lung interstitium on days 0, 2, 4, and 7 ( Fig   S1A). The percentage of D b NP366-specific CD8 T cells increased from 5% to around 50% in the WT-challenged mice, but shrank to 1.5% in the MT-challenged mice. In contrast, the D b PA224and K b PB1703-specific CD8 T cells increased from 14% to 40% in the MT-challenged mice, distinct from the modest increase of 10% to 13% observed in the WT-challenged mice. The kinetics in the airways had greater variation but followed the same pattern (Fig S1B). These data suggest the loss of D b NP366-specific CD8 T cell response might be compensated by the CD8 T cells against other epitopes.
Finally, we attempted to infer the transmission fitness of the viruses based on their replicative fitness, which is estimated by the AUC of viral kinetics. Since the relationship between transmissibility and viral loads in the respiratory tract hasn't been well established, we conducted the AUC and bootstrap analysis assuming the relationship is linear or sigmoidal (Table S1. Also

Discussion
T cell-inducing vaccines have been proposed as a potential means to lessen the burden of influenza disease due to their ability to limit viral replication and immunopathology and the high conservation of immunodominant CD8 T cell across different influenza strains [6][7][8][9][10]. Whether these vaccines may drive the evolution of cellular immune escape influenza variants has not been investigated comprehensively, partly due to the lack of empirical measures for selection pressure mediated by memory CD8 T cells. In this study, we measured the replicative fitness of wild-type PR8 Why is the selective advantage conferred by escaping the NP366-374-specific memory CD8 T cells relatively small? A likely mechanism is that the ability to evade the NP366-374-specific response is compensated by memory CD8 T cells specific for other immunodominant epitopes, namely, the PA224-233 and PB1703-711 epitopes [36,37]. This compensation was evident in B6 mice primed with x31 and challenged with either WT or MT viruses alone, where expansion CD8 T cells specific for the additional immunodominant FluPA and FluPB1 epitopes was evident only in mice challenged with MT virus (Fig S1). Overall, these data suggest that the breadth of epitope recognition by CD8 T cells enables the cellular immune response to at least partially compensate for the loss of a single immunodominant epitope. However, previous studies have shown the presentation of epitopes vary across different cell types [38], and that may also impact the advantage a virus may gain through an escaping mutation.
Interestingly, in the MHC heterozygote (H-2 b/d ) F1 mice, the advantage acquired by the MT virus started with a lower value, reached the same level as B6 mice on day 4, and leveled off thereafter. We conjectured that this difference stems from two mechanisms. First, NP366-374-specific CD8 T cells contribute to a majority of the cellular immune response to H-2 b -restricted epitopes in the lungs during a secondary influenza infection [36,37], and its loss may not be fully limited selection pressure against a single epitope. Additionally, to our best knowledge, the protection mediated by lung CD8 TRM is transient. In mice, the lung CD8 TRM immunity wanes within 6 to 7 months after initial infection [16,39,40]. Although the waning time in human remains unclear, as people getting influenza infection every 5 to 7 years [41], the selection pressure from lung CD8 TRM at individual level would be much smaller than what we observed in this study.
Combining the data with our prior modeling work [21], we suggested the conservation of CD8 T cell epitopes in influenza A viruses can be explained by the small and transient selection pressure from cellular immunity as well as the potential fitness defect due to escaping mutations.
It stands to reason that viral replication and viral transmission are linked, but exact function linking these two quantities has not been well

Measuring viral load using droplet digital PCR
RNA of virus stocks was isolated using QIAamp Viral RNA Mini Kit and stored at -80C until use.
For RNA isolation from infected mice, lungs were minced and preserved in RNALater at -80C until all the samples of the same batch were collected. Total RNA was isolated from lung homogenates using Ambion RNA Isolation Kit and the concentration was measured by Nanodrop. For reverse transcription and ddPCR, cDNA from the total RNA of lungs or viral RNA was made using Maxima RT and universal influenza primers. The cDNA samples then underwent 10-fold serial dilution and were quantified by ddPCR with probes designed to discriminate wild-type and NP366-374 mutant epitopes. The RNA copy number data were back-calculated to the RNA concentration (copy number per ng of total RNA, see Supplemental Information).

Statistical tests
The in vitro viral titer data were tested by a three-way ANOVA to account for (i) the virus strain effect, (ii) temporal effect, and (iii) batch effect. The data were then used to estimate the parameters and the 95% confidence intervals of the logistic growth model. The log-transformed in vivo viral load data were tested by Student's t-test for paired data. All the analyses were done in R 3.6.1.

Area under the curve and bootstrapping
The area under the in vivo viral growth curve (AUC) represents the total amount of virus produced over the course of infection and serves as a marker for the replicative fitness. Based on our assumption that the transmission fitness is related to replicative fitness, we log-transformed the viral load data and calculated the AUCs of the mutant and the wild type viruses. Bootstrapping was performed one million times to assess the uncertainty of estimation (see Supplemental Information). The 99% empirical bootstrap confidence intervals were calculated following basic bootstrap method [43].

Bootstrapping
For each immune setting, the bootstrapping on AUC was done by iterating the following procedure: (1) On each time point, of which n mice were measured, randomly sample n mice with replacement.
Therefore, each sampled mouse will give one WT and one MT viral load.
(2) Calculate the mean of log-transformed WT and MT viral loads, and then compute the AUCs of WT and MT with the means.
This procedure was repeated for a million times, and the bootstrapped AUCs were used to approximate the sampling distribution.

Sensitivity Analysis
To investigate the effects of link function on the estimates of selection coefficient in transmission, The constant (b) is assumed to be 0. In T1, this assumption indicates transmission does not happen when there is no virus. In T2 and T3, it means transmission does not happen when the viral load is below 1 copy/ng, which is the detection limit of our ddPCR system. We notice that the choice of coefficient (a) does not affect the estimate of selection coefficient; however, the rate (r) and midpoint (x0) of Hill function will. In Table S1 we adopted the estimates from Handel et al. [45], where r = 4.8 and x0 = 2.6, and we ran sensitivity analyses on these two parameters. were assessed in mice challenged with WT (circle) or MT (triangle) viruses.  The selection coefficients were calculated based on the whole infection course (naïve: days 0-9; others: days 0-8).